Will down scalling a 4K image to HD yeild lower noise?

[leonel escobar]

I've read on different threads about people shooting at higher ISO in 4K in low light situation in order to down scale to HD and achieve a higher light sensitivity at a lower ISO level. Can anyone confirm whether this is true? I attempted this with a C500 but as far as I could tell the noise looked exactly the same. I had set it to 3200 ISO so I could be able to see the difference. Not sure if REDs or other cameras behave the same. Thanks!

 

[Jeffery Anderson]

umm, it work on the same principal as film, but what we are talking about is the oppose effect. i.e. if you shot 8mm and blew it up to 35mm the grain would be larger and more obvious... 35mm on HD TVs the same.

with 4k, 5k, 6k digitally if you underexpose and use gain to raise the exposure levels, then you add digital noise, which can look like grain e.g. Soderberg's Knick. So, when you downsample the same principal above applies in reverse and the grain becomes smaller and less noisey.

It also helps to add a de-noiser and export with professional equipment.

PS: maybe I'm talking out of school, but I believe beside the debayer, Red just takes the signal from the sensor, where cameras from Sony and Canon compresses and denoisez the signal(which doesn't allow as much latitude in post)

 

[Jake Wilganowski]

Yes red doesn't add any noise reduction in camera, play with the free version of neat video and you will be amazed

 

[Rodrigo Prata]

The process of downscale itself (when you convert your image to a lower resolution) may takes care of some of the noise. Downscale process works by sampling a higher number of pixels and transforming then in one. So when downscaling a 4K image to a Full HD image, the most simple algorithm will sample simple groups of four pixels and average their values (in terms of color and luminace) into one. But the most complexes algorithms will not just sample groups of four pixels. They will take into account the pixels that are around in a larger are. So let's say that you are shooting a blue dress, and the footage has some noise in it, tipically, noise is green or magenta. So there are a few random pixels that are green and magenta in a area that's manly blue, the algoritym will "realize" that the random green and magenta pixels are not part of the equation and are not to be considered. Of course, most of the image is not composed by large amounts of solid colors, but the principle is more or less the same.

Noise reduction softwares work similarly, but the diference is that they are looking for the noise, and instead of just sampling the area spatially, it's sampling it temporarily as well. That means that the software will recognize when there's a noise pixel that it's there on frame, but it's not there the next one (because noise is random), and it will eliminate and substitute the pixel noise, by a pixel the same color of what is present most of the time.

Of course it's a little more complex than that, but is good way to understand.

Also, typically, noise reduction process reduce a little of the sharpness in the image, and shooting at a higher resolution then the intende final output, will help you maintain a sharp image, even with noise reduction.


I`ve shot at 6K with the Dragon (LLOLPF) for a 2K finish, and some scenes we shot at 4000 ISO and didn't feel the need to aply any noise reduction at all.